US 4106053 A Abstract A converter to translate a sampled color television video signal, consisting of four pulse-code-modulated 8-bit digital samples during each cycle of the color subcarrier, to a corresponding signal consisting of three pulse-code-modulated 8-bit digital samples during each cycle of the color subcarrier. The four 8-bit samples from present and preceeding cycles are stored in several 8-bit input registers. The 8-bit sample in one of the input registers is coupled to an output register once per cycle. Two predetermined weighted combinations of 8-bit samples in predetermined ones of the input registers are coupled to second and third output registers once per cycle.
Claims(5) 1. A converter to translate a color television video signal consisting of four digital samples during each cycle of the color subcarrier to a corresponding signal consisting of three digital samples during each cycle of the color subcarrier, comprising
a plurality of input registers connected to receive a respective plurality of successive digital samples, three output registers for three output digital samples per cycle of the color subcarrier, means to couple the contents of one of said input registers to one of said three output registers, means to weight and combine the contents of predetermined ones of said input registers and couple the result to a second one of said three output registers, means to weight and combine the contents of predetermined ones of said input register and couple the result to a third one of said three output registers, and means to couple the contents of said three output registers in sequence at the cycle rate to an output channel. 2. A converter according to claim 1 wherein said digital samples each consist of plurality of pulse code modulated bits, and each of said input registers and output registers consists of an equal plurality of parallel stages.
3. A converter according to claim 2 wherein said input registers comprise four multi-bit shift registers each receiving and shifting in parallel the plurality of bits of a respective one of the four digital samples at the cycle rate, whereby stages of the shift register contain digital samples from a present cycle and at least one preceding cycle.
4. A converter according to claim 1 which each of said means to weight and combine the contents of predetermined ones of said input registers comprises arithmetic shift registers in which a digital sample is shifted to the right to divide the digital number by the number of shifts.
5. A converter according to claim 4 wherein each of said means to weight and combine the contents of predetermined ones of said input registers also comprises adder means for combining the outputs of said arithmetic shift registers.
Description This invention relates to converters for translating a sampled color television video digital signal consisting of a given number of digital samples during each cycle of the color subcarrier, to a digital signal having a different number of digital samples during each cycle of the color subcarrier. A color television video signal in the usual analog form can be translated to a digital form by making four digital samples per cycle of the subcarrier, or by making three digital samples per cycle of the subcarrier. Since the color subcarrier frequency is 3.58 MHz, the four samples per cycle occur at a rate of 14.32 MHz and provide a very high quality digital signal. Three samples per cycle occur at a lower rate of 10.74 MHz and provide a digital signal which can be more economically transmitted over a narrower band channel. The present invention is useful, for example, to convert from four digital samples per subcarrier cycle to three digital samples per subcarrier cycle. The latter signal is more suitable for economical transmission to and from a satellite in earth orbit. A known method of making the conversion is to use a digital-to-analog converter to convert the signal with four samples per cycle to an analog video signal, and then use an analog-to-digital converter to convert the analog signal to a digital signal with three samples per cycle. This known method introduces distortion which somewhat degrades the quality of the television signal. According to an example of the invention, a digital-to-digital conversion is made from four samples per color subcarrier cycle to three samples per subcarrier cycle by giving predetermined weights to predetermined digital samples occurring during present and preceding cycles, and by making predetermined combinations of the weighted samples to form three output samples during each cycle of the color subcarrier. In the drawing: FIG. 1 is a block diagram of a system constructed according to the teachings of the invention; FIG. 2 is a schematic detailed diagram of one portion of FIG. 1; FIG. 3 is a timing diagram which will be referred to in describing the operation of FIG. 1; FIG. 4 is a table showing the nomenclature applied to successively-received digital samples. FIG. 5 is a block diagram of an alternative construction for the shift registers in FIG. 1; FIG. 6 is a block diagram of an arithmetic unit useful in the system of FIG. 1; FIG. 7 is a chart of filter characteristics which will be referred to in describing how weighting constants are derived; and FIG. 8 is a chart of calculated variations in signal/distortion ratio with frequency when described approximate coefficients are used. In FIG. 1, four serial digital samples per color subcarrier cycle from a source 8 are coupled over an 8-conductor bus 9 to inputs of four three-stage 8-bit-parallel shift registers 10, 11, 12 and 13. Three stage shift register 10 operates at time t At a time t At this time t FIG. 4 shows the nomenclature used for successively-received digital samples. The first sample x It will be noted that some of the registers 20 through 29 in the four three-stage shift registers 10 through 13 are not used in the specific example illustrated in FIG. 1. A more economical implementation including simply seven storage registers and transfer gates for the four 8-bit samples is shown in FIG. 5. The operation of FIG. 5 is clear when one begins with acceptance of the first sample x The arithmetic unit 32 in FIG. 1 which will be described in connection with FIG. 6, produces three 8-bit digital signal samples which are present at time t The arithmetic unit shown in detail in FIG. 6 includes a connection 50 from input terminal x The y The particular numbers by which the four input samples are divided, and the combinations in which the divided samples are added, and the sums combined in the arithmetic unit of FIG. 6 are determined by a technique for determining the sample values at a sampling rate equal to three times the color subcarrier. The description of the technique is divided into three sections. In the first section the constructed sample will be represented as a weighted sum of the original samples, the weightings being determined by a filter. Several filters will be investigated, and the one which uses the fewest of the original samples to construct the new sample will be chosen. In the second section, the signal-to-distortion ratio of the resulting signal will be determined. To determine this ratio, the spectrum of a typical TV signal will be used. The distortion will be weighted to reflect the expected severity of distortion in different segments of the frequency band. In the third section, the weighting factors will be modified, so that the circuit can be implemented using adders rather than dividers, and then perturbed to obtain a local maximum in the signal-to-distortion ratio. The resulting circuit will use between one and six old sample values to construct a new sample. It will provide a 74 dB signal-to-weighted distortion ratio in a 4.5 MHz band. I. choice of Weighting Functions A signal X(t), sampled once every T seconds, results in a signal ##EQU1## The frequency spectrum of the resulting signal contains the original frequency spectrum plus displaced versions of that spectrum. To reconstruct the original signal the sampled signal is filtered by g(t), which eliminates the displaced spectra. The resulting signal is: ##EQU2## If this signal is sampled every 4/3 T seconds, ##EQU3## The new samples are thus represented as a weighted sum of the previous samples. The conversion between sampling rates can be effected by multiplying the old samples by constants, determined by a filter, and summing. Due to the symmetry of the problem, the same weighting functions will be used to determine every third one of the constructed samples. Separate circuits must be constructed to determine y(0), y (4/3 T), and y (8/3 T), and this circuitry can be used to determine all other samples. To realize this circuit, the summation in equation 1 must be truncated to some finite value. This value depends on the filter chosen and the amount of distortion which can be tolerated. Initially, all weighting coefficients less than 1/512 were ignored. The reason for choosing this level is that the sampled signal is quantized to 256 levels. If the weighting factor is less than 1/512, it cannot add more than 1/2 a quantization level to the constructed sample, and will most likely not effect the final value. Using this criterion, over 200 of the original samples are required to construct each new sample, if a rectangular low pass filter is used to determine the weighting coefficients. This means the conversion circuit would require over 600 multipliers and adders. To reduce the complexity of this circuit filters with more gradual skirts in the frequency domain were considered. Among the filters considered were those with triangular, raised-cosine and gaussian skirts. The filters are flat in the frequency range 0→f If f If f If f Taking these phenomena into account, and assuming the signal energy in the range between 4.2 and 6 MHz is small in comparison to the energy between 0 and 4.2 MHz, skirts were designed to fill the band between 4.2 and 10.12 MHz. A raised-cosine skirt in this range reduced the number of significant weighting coefficients from over 200, for the low-pass filter, to 21. A filter with a gaussian skirt was investigated. The response of this filter was down 50 dB at 10.74 MHz, where the shifted version of the color subcarrier occurs. This requirement made this filter look more like a low-pass filter than the raised cosine filter, and, increased the number of significant weighting coefficients. It is possible to obtain faster convergence in the time domain using gaussian filters, than it is using raised cosine filters. This benefit was not realized in this instance because of the severity of the constraint on attenuation at a high frequency. Various arrangements of gaussian functions were investigated to try to obtain this potential improvement, and, it was found that the convergence could be made faster by defining a split gaussian filter, as shown in FIG. 5, and as defined by the following equations: ##EQU4## The following constraints were placed on the design of this filter: Both the function and its first derivative are continuous at F1. The attenuation is down 53 dB at 3×3.58 MHz, where the displaced color subcarrier occurs. The signal-to-distortion is 53 dB at 3.58 MHz. The constraints set k1 = EXP [η Ii. signal/Distortion The filtering techniques introduce distortion into the signal spectrum, because of the phenomena described in Section I, and the series truncation performed. This distortion will be calculated. The constructed signal is: ##EQU6## where F
______________________________________ ##STR1##
______________________________________ It is noted that the relationship between y(k(4/3)T) and the original samples X(NT), is the same as the relationship between y(k+3j) (4/3)T) and the original samples X((N+4j)T). The constructed signal can therefore be represented as: ##EQU7## where ##EQU8## That is, the F Every third one of the constructed samples lines up with one of the original samples. It was found that when the filter has the type of symmetry, about 2·3.58 MHz displayed by the split gaussian and raised cosine filters, the constructed sample equals the original sample it lines up with. Furthermore, it was noted that the coefficient multiplying the sample at T-iT, when determining the sample at 4/3 T, is the same as the coefficient multiplying the sample at 3T+iT, when determining the sample at 8/3 T. This is expected from the symmetry of the samples in the time domain. The constructed signal can therefore be represented as: ##EQU9## where, ##EQU10## Taking the fourier transform of the constructed signal, and simplifying, one obtains, ##EQU11## where f To construct the estimated signal from Y(f), it is passed through a low pass filter with a cut-off at 4.5 MHz. The distortion component is: ##EQU12## To calculate the distortion it is necessary to assume a signal spectrum. The spectrum used in these calculations is taken from D. G. Fink, Television Engineering Handbook, McGraw-Hill Book Company, 1957, pg. 10-22. According to Fink, the spectra of a set of 27 slides provided by the Eastman Kodak Company for the National Television System Committee, has the following characteristics: The luminance component decreases in amplitude at a rate of approximately 6 dB per octave in the range from 0.0157 to 0.75 MHz, 10 dB per octave from 0.75 to 4 MHz, and 12 dB per octave above 4 MHz. The chrominance components decrease at about 7 dB per octave in the range from 15.7 to 400 kHz, 5 dB per octave from 0.4 to 1.5 MHz on the side toward the picture carrier, and 9 dB per octave from 0.4 to 1.5 MHz on the side away from the picture carrier. The ratio of the peak luminance amplitude to the peak chrominance amplitude is about 5 to 1. For a particular set of weighting coefficients, this calculation provides a distortion spectrum for the band of the TV signal. It has been suggested that distortion in all regions of the band are not equally objectionable. In particular, it is felt that distortion in the low frequency region of the band and distortion in the region near the color subcarrier is much more objectionable than distortion near the 2 MHz region, and distortion near the high end of the band. While a quantitative measure of objectionability for this type of distortion is not available, several weighting functions for white noise have been established. The CCIR noise weighting function: ##EQU13## F F F emphasizes distortion in the regions in which it should be most objectionable. This function was uesd to weight the distortion power spectrum. Iii. modification of Multiplication Coefficients The significant multiplicative coefficients, obtained from the split gaussian filter, were: A A A A A A A A A To implement this circuit in hardware it is desirable to represent A The following procedure was used to obtain a set of A A A A A A A A A A The peak-to-peak signal power-to-mean square weighted noise power was calculated as follows. The mean square signal power was calculated by numerically integrating the square of the amplitude signal spectrum, between 0 and 4.5 MHz. Assuming the signal levels to be uniformly distributed between +V and -V, S The procedure was repeated for small intervals in the band of the signal. The signal-to-weighted distortion ratio at different frequencies in the band is plotted in FIG. 8. Using the results of the foregoing calculations, seven digital samples x Patent Citations
Non-Patent Citations
Referenced by
Classifications
Rotate |